The AI Nature of the Firm: From Agents to Collective Intelligence

Summary

Ronald Coase’s 1937 theory of the firm explains why people form organizations: markets coordinate broad activity, but firms are better suited to detailed decisions, planning, and the aggregation of limited human resources. Cameron Gordon argues that the same pressures may shape the next stage of AI. Interest in Moltbook, a Reddit-like environment where decentralized AI agents interact through message boards, has highlighted a form of agent collective, although the author describes it as more chaotic than a mature organization. At the AAAI conference in Singapore, many papers focused on autonomous agents using tools, running scientific workflows, and operating in domains such as finance and health. Gordon notes that most work still centers on individual agents or small interactions, while human organizations can aggregate information, coordinate specialized expertise, and act at a scale no individual can match. He connects this prospect to Herbert Simon’s work on administrative behavior, corporate decision-making, and management hierarchies as information filters. As multi-agent systems expand from dozens to thousands of agents, he expects similar organizational pressures to reappear in AI research. The essay suggests that economists and management scholars could contribute to this shift through mechanism design and collective decision theory, and speculates that 2026 and 2027 may mark the beginning of the “AI corporation” era.